Energy Consumption in Data Analysis for On-board and Distributed Applications

Energy consumption is an important issue in the growing number of data mining and machine learning applications for battery-powered embedded and mobile devices. It plays a critical role in determining the capabilities of a broad range of applications such as space probes with onboard scientific missions, PDA-based monitoring of remote data streams, event detection in sensor networks comprised of battery-powered data sensors and light-weight data processing nodes. This paper presents an experimental investigation of the energy consumption characteristics of different data analysis techniques. The paper compares the energy consumption characteristics of common data analysis operations on-board a mobile device with the energy necessary to send the same amount of data over wireless networks to a remote machine for analysis. It benchmarks the performance and points out that the energy consumption for transmitting data over low bandwidth lossy wireless channels often supersedes that for the operations on-board a light-weight compute node.